The Convolution Operation — Checkpoint Quiz

A is correct. A convolution kernel is a small grid of weights. At each position, it multiplies elementwise with the underlying patch of the image and sums the result, producing a single value that's high when that local patch matches the pattern the filter has learned to detect.

A is correct. The same small filter (e.g. 3x3) is reused at every position across the image, rather than learning an independent weight for every pixel-to-output connection. This parameter sharing is what makes convolutions so much more efficient than full connectivity for image-like data.

A is correct. Without padding, each convolution shrinks the spatial dimensions slightly since the filter can't center on edge pixels. Adding a border of (usually zero) padding lets the output preserve the input's height and width if desired.